-
Why I show my machine: an AI, a library and a news engine
My own AI brief once stated a wrong cause for the Microsoft outage. The closing post of the series: how I work with my AI, how claims are graded documented/reported/inferred, why LSN News counts voices instead of labelling truth, and the failure modes.
-
Who checks the fact-checkers? How it works, who pays, its record
In 91% of the LSN News stories with a fact-check, my engine grouped no other outlet's article with it. How fact-checking works, who pays (Meta fell from 45.5% to 34.3% of revenue), its documented misses and strengths, and how to read a fact-check like any article.
-
Everyday fakes: old photos, cut videos, fake quotes, satire
An AI picture of Italy's foreign minister in a MAGA cap ran in 46 articles before it was deleted. The five shapes of everyday fakes (wrong-context photos, cut videos, fake quotes, satire, synthetic faces), cases from every side, and quick tests.
-
Check a claim yourself in 10 minutes: a practical toolkit
52% of 3,446 US students took a Russian ballot-stuffing video as strong evidence of US fraud. A timed 10-minute routine to check any claim: primary sources, lateral reading, reverse image search, archives and counting independent voices.
-
What censorship looks like in 2026: from removal to quiet demotion
Censorship in 2026 is less often a ban and more often a ranking change, a lost ad or a keyboard that won't finish your phrase. The six forms, documented cases from every side, and how to tell which one is happening to you.
-
Say it again: repetition and the illusory truth effect
Hearing a claim again makes it feel truer, even when you know better. From Hasher's 1977 experiment to a headline copied on 362 sites in six hours, how repetition, slogans and repeated phrases work, and how to resist.
-
Counting voices, not trusting labels: how LSN News measures it
One headline ran on 362 sites of one radio company: 384 outlets, 5 independent voices. How LSN News counts voices and finds outlet networks (194 networks, 1,597 sites), why I count instead of labelling sources, and what counting cannot do.
-
What propaganda cannot do: the limits of persuasion
49 field experiments put the average effect of campaign contact on votes at zero, and the 2016 troll tweets show no measurable effect. Yet radio helped drive genocide in Rwanda. What propaganda can and cannot do, and why fearing it too much is a trap.
-
Information hygiene: a daily routine that takes 15 minutes
In one day, 191 big stories in my news engine averaged 43 outlets but only 17 independent voices. A 15-minute daily routine against the drip: notifications off, slow news, count voices not outlets, three sources, and one question before you share.
-
How to talk to someone who believes something you don't
A 10-minute conversation with 501 voters shifted attitudes more than 14 years of social change, and it lasted three months. What deep canvassing, motivational interviewing and street epistemology show about talking across a disagreement.
-
Outrage pays: what recommendation systems do to attention
Facebook once weighted an 'angry' reaction at five likes. Studies show moral outrage and attacks on the other side spread further. But the evidence on filter bubbles is mixed, and a big 2020 experiment found chronological feeds did not reduce polarisation.
-
The gatekeepers you don't see: search ranking and autocomplete
Ranking decides what you see first. A 2015 study said biased search results can shift undecided voters by 20%; later vote-count claims are contested. Real users' clicks matter more than the algorithm. And on LSN News one story ranked 1st in Italian and 70th in Hindi.